Bodhan AI, a Centre of Excellence in AI for Education incubated at the Indian Institute of Technology Madras (IIT Madras), has launched a suite of open foundational artificial intelligence models for Indian languages in collaboration with NVIDIA.
The models have been developed with AI4Bharat, the Indian language technology research initiative at IIT Madras. They are intended to support multilingual applications in education and other public-interest areas by providing developers, researchers, startups and government institutions with a common set of language technology tools.
Four models for Indian language technology
The suite covers four core capabilities:
Indic-Transcribe: Speech recognition and speech-to-text conversion
Indic-Speak: Text-to-speech and speech generation
Indic-Translate: Machine translation across Indian languages
Indic-OCR: Optical character recognition for extracting text from printed and handwritten documents
The models are being released as open-weight models, allowing users to access, adapt and build applications using them. Hosted APIs are also available through Bodhan AI’s digital infrastructure.
According to Bodhan AI, Indic-Transcribe supports more than 25 Indian languages and English, including regional accents, dialects and code-switching. Indic-Translate supports translation between English and the 22 scheduled Indian languages, while Indic-Speak supports speech generation in Indian languages and English. Indic-OCR is designed to process printed and handwritten text, including equations and tables.
Developed using NVIDIA technology
The models were developed using the NVIDIA NeMo framework. The collaboration also includes the use of NVIDIA Nemotron technology for speech recognition, along with NVIDIA TensorRT-LLM and vLLM microservices for inference.
Bodhan AI and NVIDIA are also working on datasets, training methods and evaluation processes for future foundational models focused on Indian languages.
Prof Mitesh Khapra, principal investigator at Bodhan AI and AI4Bharat, said the partnership would help make advanced language technology available to a wider developer and education ecosystem.
Focus on education and public digital infrastructure
The models form part of the proposed Bharat EduAI Stack, which Bodhan AI describes as a sovereign digital public infrastructure for education. The initiative aims to provide a common technology layer for education-focused AI applications instead of requiring every institution or organisation to develop the same capabilities independently.
Potential applications include multilingual learning resources, speech-based learning tools, translation, document processing, accessibility services and teacher-support systems. Bodhan AI has said that educational applications using the models will remain free for learners, teachers and partnering state governments.
Bodhan AI’s broader research agenda includes speech, language, vision and learning technologies across more than 22 Indian languages. Its website lists applications such as child speech assessment, educational OCR, multilingual speech recognition and text-to-speech tools designed for classroom use.
The open release is intended to allow edtech companies, universities, researchers and government partners to build and customise multilingual AI applications for Indian users.